How it works

From scattered documents
to a living specification.

Eight steps of one loop. Each with proof, history and a clear status — what works today, and what's coming next.

01Available

Import. From every source.

Upload specs, emails and call transcripts. Specky extracts the requirements and instantly points out where the gaps and contradictions between documents are.

  • DOCX, PDF, Markdown — many files at once
  • Requirements grouped by section
  • Conflicts between sources caught automatically
project_alpha · 3 sourcesextraction
📄
spec.md
📧
email Mar 14
🎙️
call May 28
47 requirements 3 gaps 1 contradiction
02Available

Every requirement has proof.

You don't have to take AI's word for it. Click a requirement and see the exact part of the document it came from — with a verbatim quote and line numbers.

  • Linked to specific source lines
  • A verbatim quote, not a paraphrase
  • When AI can't quote it — it says so directly
email_mar14.emlrequirement #12
2From: anna@client.com
3Subject: change request
4Please export to XLSX
5instead of CSV. Reason: Excel.
6Best regards, Anna Kowalska
Requirement #12 → 📧 email Mar 14 · lines 4–5
03Available

AI proposes. You decide.

Specky suggests questions for the client and concrete change proposals. You accept, edit or reject. Clarifications add acceptance criteria — nothing overwrites previous decisions.

  • A proposal as a ready "was → now" change
  • Acceptance criteria add up on the requirement
  • Every decision saved with context
AI proposes a change · requirement #12Available
Export reports to XLSX
was: export to CSV
+ now: export to XLSX
✓ Accept
Edit
Reject
acceptance criteria #12
XLSX format
separator ";"
max 100k rows
+ range: last 3 months
04Available

Walk into the meeting with a ready list of questions.

Questions marked "for consultation" automatically build an agenda. After the meeting you paste the decisions, and Specky propagates them to the right requirements.

  • An agenda from open questions — automatically
  • Pasted decisions → change proposals in the spec
  • Two statuses: discussion and implementation
agenda · meeting May 282 / 3 closed
To discuss
Which fields to filter in reports?
After discussion
Notifications — email or push?
→ email only · propagates to §3
Export limit?
→ 100k rows · propagates to §4
05Available listening in progress

The spec grows with the project.

Every new decision — from a pasted email, an uploaded transcript, and soon straight from Slack — Specky links to context and proposes what to update. It remembers history and never asks twice.

  • New decision → change proposal in a section
  • Knows the history — recognizes what's already decided
  • Versioning: you see every change
event stream · spec.md v3.1+2 changes
📧
Email: "filters sorted by priority"
new decision → §5 · spec v3.2
💬
Slack: "client asks for XLSX"
already known → question from Mar 14 · skipped
🎙️
Transcript: "range of the last 3 months"
new requirement → §4 · spec v3.3
06Available

You know how far you are from ready.

Spec readiness is a single 0–100% score. Every resolved ambiguity raises it — you see progress and know when the spec is ready to hand off.

  • A deterministic score, not guesswork
  • A "+N%" notification after every decision
  • A trend over time — whether the project is maturing
spec readiness7-day trend
31%
✨ Spec readiness +12% after a decision
07Available

No document? Start from an idea.

Concept mode leads a conversation about your idea — problem, user, risks — and catches candidate requirements. The Canvas organizes it all before you move to the specification.

  • An ideation chat instead of a blank page
  • "That sounds like a requirement — add it?"
  • Canvas: problem, market, risks, options
concept mode · ideation chatCanvas
What problem do you want to solve?
Generating reports takes 2h a day.
✨ That sounds like a requirement + Add as requirement
Problem
Manual reports — 2h/day
User
Client's accounting
Risks
Date formats PL vs EN
Options
Export CSV / XLSX
08Available

The finished spec drives the work ahead.

Download the clean spec as MD or PDF — for the team or the client. Or connect it to Claude via the connector: AI reads the requirements and criteria and proposes code with full context.

  • Export MD / PDF + version snapshots
  • MCP connector — Specky as a source for Claude
  • AI proposes, the decision is still yours
export and connectorspec.md v3.3
⬇ Markdown
⬇ PDF
🤖 Claude — connectorvia MCP
Reading 47 requirements + acceptance criteria → I propose implementing the /export endpoint per spec §4.

See the whole loop on your own specification.